The use of unmanned aerial vehicles (UAV) for image based surface reconstruction has gained enormous popularity in the last decade, not only in the study field of geomorphology, as technological improvements continue steadily. UAV Systems (UAS) have shown to be a cost-effective short- and close-range remote sensing alternative to classical manned aerial photogrammetry or LiDAR surveys. As geomorphic investigations concentrate on a multitude of purposes and processes different demands on accuracy and resolution arise.
This study focuses on the error assessment of 3D point clouds (PCs) derived from high resolution image sets of four different UAV-systems: DJI Mavic Pro (MAV), DJI Phantom 4 Pro+ (P4P+), DJI Inspire 2 (I2) and Airborne-Robotics XR6 with Sony Alpha 6000 (XR6) with fixed 19 mm Sigma objective. We defined three classes to categorize the systems: “low-cost” ( 15000 €). In particular, we analyzed the influence of different flight patterns (parallel axes, cross-grid, cross-grid + oblique images) on the accuracy of derived PCs for each system. External orientation was achieved by the establishment of 25 ground control points (GCPs) measured with a Global Navigation Satellite System (real time kinematic mode). Additionally, 156 independent validation points were measured in order to calculate vertical differences between the reconstructed surfaces and GNSS validation points. We used a standard SfM-MVS workflow (Photoscan Pro 1.4.2 by Agisoft LLC) to transform the image sets into 3D PCs.
Semi-variograms were calculated to address the statistical and spatial independency of errors of the single PCs. The preliminary results show that, both, point accuracy and spatial independency of errors differs strongly with regard to flight patterns and different UAS.